Finout Blog Archive

AWS FinOps Agent: How It Works, Setup, Pricing, and Limits

Written by Finout Writing Team | Nov 17, 2026, 9:40:00 AM

Key Takeaways

  • What it is: AWS FinOps Agent is AWS's preview AI agent that answers AWS cost questions, investigates anomalies, and runs scheduled FinOps reports.
  • How it works: It reads from Cost Explorer, Cost Anomaly Detection, Cost Optimization Hub, and Compute Optimizer, plus CloudTrail, AWS's change log, to trace spikes.
  • What it changes: Read-only by default against AWS Billing and Cost Management products, it takes write actions, such as creating Jira tickets or posting to Slack, only through integrations you configure.
  • Who it fits: Based on AWS's documentation, every data source is an AWS service; Kubernetes, SaaS, and direct AI bills need wider allocation.

What Is AWS FinOps Agent?

AWS FinOps Agent is an AWS AI agent that answers cost questions, investigates cost anomalies, surfaces savings recommendations, and runs scheduled FinOps workflows. It launched in preview on June 9, 2026, and AWS calls it a frontier agent. It uses Amazon Bedrock foundation models.

The AWS preview announcement describes it this way: "Today, AWS announces the preview of AWS FinOps Agent, a frontier agent for FinOps practitioners and engineering teams that answers cost questions, surfaces optimization opportunities, automatically investigates cost anomalies, and runs recurring FinOps workflows on a schedule you define." AWS positions the agent as complementary to its existing cost services, and it reads from them instead of replacing them.

AWS FinOps Agent at a Glance

The table compares each AWS service the agent draws on, what that service does on its own, and what the agent adds.

AWS service What it does on its own What the agent adds
Cost Explorer Shows cost and usage data in charts and reports Answers plain-language questions and builds reports from that data
Cost Anomaly Detection Flags unusual cost spikes Investigates each flagged spike
Cost Optimization Hub and Compute Optimizer Recommend rightsizing, idle resource cleanup, and Savings Plans Summarizes recommendations and routes them to Jira
CloudTrail Records who changed what, and when Matches those changes to cost spikes

For how these services work without the agent, see these AWS cost management practices.

How Does AWS FinOps Agent Work?

The agent sits on top of AWS's own cost services and reads the same data your FinOps team already uses. According to the AWS launch blog post, "It draws on AWS Cost Explorer, Cost Anomaly Detection, Cost Optimization Hub, and Compute Optimizer, so its answers reflect the same data your central FinOps team relies on."

Each agent lives in your AWS account and uses its own IAM roles, which are permission sets that control what it can read and do. A management account agent, built in the account that owns billing for your AWS Organization, covers every account; member account agents see only their own.

Cost Questions in Plain Language

You ask the agent questions in its web app the way you'd ask a colleague. AWS's own example is "Why did my AWS cost go up last month?" The answer names the cost changes, the services behind them, and the usage that drove them.

The agent can break spend down by service, account, Region, or tag, and compare time periods. Specific prompts get specific answers. AWS suggests prompts such as "List the top 10 cost drivers last month and group them by Region."

Anomaly Investigation With CloudTrail

CloudTrail is AWS's log of who changed what, and when, across your AWS environment. When Cost Anomaly Detection flags a spike, the agent compares the spike with CloudTrail events and identifies the change that likely drove it.

The agent then writes a summary with the likely root cause and the responsible owner. You can add a dollar threshold to the automation prompt, for example only investigating anomalies above $1,000, which keeps small changes out of your queue.

AWS's own wording is "likely root cause," and that word matters. Treat each summary as a strong lead for the resource owner to confirm. Detection without investigation is noise, and an investigation without a confirmed owner is still a guess.

Scheduled Reports and Automations

The agent runs work three ways: on demand, on a recurring schedule such as daily, weekly, or monthly, or when an anomaly event arrives. Schedules suit recurring reviews, and event triggers suit anomaly response.

Reports cover more than a cost total. The AWS FinOps Agent product page says, "The agent creates HTML, PDF, and PPTX reports with cost summaries, breakdowns by service, account, Region, or tag, trend analyses, month-over-month comparisons, and forecast projections." If you don't pick a format, you get HTML.

Automations are written as plain instructions. AWS's example asks the agent to "listen for AWS Cost Anomaly Detection events, investigate each anomaly for root cause, and post the findings to the #finops-anomalies Slack channel."

Context Files and Memory

A context file is a document you upload to tell the agent how your organization works. Examples include account-to-owner mappings, team definitions, tagging conventions, and review cadences.

The agent uses these files to resolve a question like "What's the cost of Team X?" to the accounts that team owns. It also remembers your preferences and corrections across sessions.

The mapping is a file you build and keep current, and an account-level mapping can only answer at the account level. If two teams share an account, the file can't split their costs, which is where cloud cost allocation methods come in.

What Can AWS FinOps Agent Change in Your Environment?

By default, the agent reads your cost data and leaves it alone. The AWS FinOps Agent FAQs state, "By default, the agent operates in read-only mode against AWS Billing and Cost Management products."

  • Write actions: It creates Jira tickets or posts to Slack only when you've configured that integration and it's running a task you asked for.
  • Jira: An admin installs it as a Forge app, Atlassian's app framework, and tickets go only to the Jira space keys you connect.
  • Slack: Slack is delivery-only in preview, meaning the agent posts findings and reports, but engineers can't start conversations with it there.
  • Audit trail: CloudTrail captures all agent activity in the hosting AWS account, giving your security team a record to review.

Before you connect Jira, decide which spaces should receive automated tickets. A ticket in a queue nobody watches is the same as no ticket.

How to Set Up AWS FinOps Agent

Decide between a management account and a member account before you start, because that choice sets what the agent can see. AWS's getting-started list runs in this order:

  1. Create the agent. Sign in to the AWS Management Console, switch to US East (N. Virginia), and create an agent in the AWS FinOps Agent console.
  2. Set up IAM roles. Complete the one-click setup that provisions the roles the agent uses to read cost, usage, and operational data.
  3. Connect Jira and Slack (optional). Add the Jira space keys and Slack channels the agent can post to.
  4. Confirm the configuration. Review the agent's settings in the console.
  5. Open the web application. Launch the agent's web app from the console page.
  6. Upload context (optional but recommended). Add an account-to-owner mapping and rules such as known exceptions and review cadence.
  7. Run a first query. Ask a plain-language question about last month's spend and check the answer against what you already know.
  8. Set up an anomaly automation. Ask the agent to listen for Cost Anomaly Detection events, investigate each one, and post findings to Slack.

Step 8 depends on anomaly monitors already existing in Cost Anomaly Detection. If you have none, the automation has nothing to listen for. Build the mapping in step 6 before step 7, because team-level answers depend on it.

AWS FinOps Agent Availability and Pricing

The agent runs in one Region, but its data reaches much further. AWS's Region and pricing details state, "AWS FinOps Agent (preview) is available in the US East (N. Virginia) Region and includes cost and usage data covering all AWS Regions, except AWS GovCloud (US) Regions and AWS China (Beijing and Ningxia) Regions." Workloads in Tokyo or Frankfurt still show up in its answers.

AWS's launch blog says the agent costs nothing during the preview, with a monthly usage limit. Standard charges apply for other AWS services you use with it. Check the console before you build a budget around it, because preview terms can change.

Where AWS FinOps Agent's Scope Ends

Based on AWS's documentation, every data source the agent uses is an AWS service. That's our reading of the docs, not an AWS statement, and it defines what the agent can see.

  • Other clouds: Azure, Google Cloud, and OCI bills aren't among its documented sources.
  • Kubernetes: Shared clusters bill as nodes, and splitting that cost by team or namespace isn't among its documented capabilities.
  • SaaS and direct AI providers: OpenAI or Anthropic invoices paid outside AWS, plus Datadog, Snowflake, and similar tools, sit outside its data.
  • Trigger dependency: Event-based investigations start only when Cost Anomaly Detection fires.
  • Ownership dependency: Team-level answers depend on the mapping files you build and keep current.

Our breakdown of when anomaly detection misses AI charges covers an April 2026 Amazon Bedrock case. Cost Anomaly Detection never fired on $30,141.33 in Anthropic Claude charges billed through AWS Marketplace. With no alert, an event-triggered agent has nothing to investigate.

Spend beyond public cloud is now common for FinOps teams. In the FinOps Foundation's State of FinOps 2026 report, "90% of respondents now manage SaaS or have plans to (up from 65% in 2025)."

The same report found "AI management has become nearly universal at 98% (up from 63%)." If your team manages that kind of spend, part of it sits outside the agent's documented data.

What Does This Mean in Practice?

If your spend is almost all AWS and accounts map cleanly to teams, start with the agent for anomaly triage and recurring reports. Its data already matches what you manage, and the preview costs nothing within its usage limit.

If teams share accounts or clusters, build the ownership mapping before you rely on the agent's answers. An account-level file can't split shared spend, and findings will reach the wrong owner.

If you run Kubernetes, other clouds, SaaS, or direct AI provider contracts, prioritize allocation that spans all of it. A cost the agent can't see never reaches an owner. The same signals you've outgrown Cost Explorer apply here, since the agent reads the same data.

Finout is an enterprise-grade FinOps platform for cloud and AI spend, and its customer PandaDoc hit this limit with native tools. Ivan Polonevich, DevOps Team Lead at PandaDoc, said: "The native tools like Cost Explorer and Kubecost didn't allow us to create a proper dashboard with a combination of AWS and Kubernetes costs - but Finout does this easily." The case study shows PandaDoc's AWS and Kubernetes view.

Whichever path fits, ask one question next. What share of your bill can the agent see, and who owns the rest?

How Finout Brings Agentic FinOps to Every Cloud and AI Bill

Finout applies one agentic workflow across cloud, Kubernetes, SaaS, and AI spend, using one set of allocation rules. Finout's AI offerings cover both AI spend and agent-driven FinOps work.

FinOps for AI brings OpenAI, Anthropic, and Cursor costs into Finout the same way as cloud spend, including cost per token. Billy answers cost questions in plain language on live Finout data, and runs read-only.

FinOps Agents handle detection and investigation across your spend. Detection Agent scans cloud, Kubernetes, AI, and SaaS spend for waste and anomalies, using your own definitions of waste. Investigation Agent traces each finding to its root cause, blast radius, and accountable owner. Humans approve every destructive action, and reversible actions run only under policy you configure, as explained in how Finout's agents handle approvals.

FinOps for Agents connects your own agents to the same governed data through an MCP server, an API, and a Data Exporter. MCP, the Model Context Protocol, is an open standard that lets AI tools such as Claude or Cursor query data sources.

Underneath, the MegaBill unifies AWS, Azure, GCP, OCI, Kubernetes, SaaS, and AI costs in one governed data layer. Virtual Tags allocate tagged and untagged spend to teams, products, and owners without code changes, giving every finding an accountable owner.

Alchemy identified and addressed cost inefficiencies 90% faster using Finout's Anomaly Detection. The details are in Alchemy's anomaly response results.

To see agent findings reach the right owner on every bill, Book a demo.